Experimental evaluation of a cooperative kernel-based approach for robust spectrum sensing

The cognitive radio (CR) paradigm relies on accurate spectrum sensing schemes. These schemes can benefit from spatial or multiuser diversity, though they need to be able to work under non-ideal scenarios, such as those affected by impairments or interferences. In this paper, a cooperative spectrum sensing (CSS) technique is experimentally evaluated by means of a CR testbed. The proposed method is based on kernel canonical correlation analysis (KCCA), which is performed at the fusion center (FC) in a first cooperative stage. In particular, the FC extracts the local test statistics for the different sensors, without the need of any training signal or labeled data, after which the sensors can operate in a completely autonomous manner. The experiments are conducted on a cognitive radio platform composed of several Universal Radio Peripheral (USRP) nodes, and the measurements show that our scheme is able to learn the surrounding environment by exploiting only the nonlinear correlation among the signals at each secondary user (SU). Moreover, we illustrate the performance of the proposed technique in the presence of local interferers, where we can appreciate a significant performance gain over a conventional energy-based cooperative detector.